Values Of Computer-aided Diagnosis System In MRI For Small Breast Masses

Part ⅠValues of Kinetic Features Measured by MRI-CAD for Small Breast Masses[Purpose]:To investigate the value of kinetic features of small breast masses measured by computer-aided diagnosis (CAD) software for MRI.[Materials and methods]:Eighty-four small mass-like breast lesions(≤2.0 cm) seen on MRI were reviewed retrospectively. The CAD variables compared for benign and malignant lesions included initial phase peak enhancement, the curve-type distribution percentage and delayed phase enhancement categorized by single most suspicious curve. Region of interest (ROI) were drawn manually within the lesions to calculate the time intensity curves (TIC). Both methods for differentiation of benign and malignant lesions were compared.[Results]:There were 38 benign and 46 malignant small breast masses. Initial phase peak enhancement mean values of benign and malignant lesions were 230.54±83.46% (92%-442%) and 257.72±78.93%(89%-448%). There was no significant difference between benign and malignant lesions in initial peak enhancement mean value (P= 0.138). Washout, plateau and persistent voxel percentage median values measured in malignant lesions were 15.15%,20.00% and 61.50% separately. Washout, plateau and persistent voxel percentage median values measured in benign lesions were 1.06%, 10.08% and 87.00%. There was significant difference between benign and malignant small masses in the curve-type distribution percentages (P<0.001). If persistent voxel percentages less than 81.5% in a lesion were considered malignant, sensitivity of 89.1% and specificity of 60.0% were reached. Delayed phase enhancement categorized by single most suspicious type of kinetics (any washout> any plateau>any persistent) for benign and malignant lesions were 24,3,8 and 44,2,0 respectively. If a threshold of 100% initial enhancement and delayed washout were considered malignant, sensitivity of 95.7%, specificity of 40.0%, and an area under the cure on ROC analysis of 0.678 (95%CI:0.555-0.801, P=0.006) were obtained. Persistent voxel percentages measured by MRI-CAD demonstrated highest diagnostic accuracy (AUC=0.798,95%CI: 0.701-0.895, P<0.001). If persistent voxel percentages less than 81.5% in a lesion were considered malignant, sensitivity of 89.1% and specificity of 60.0% were reached, and CAD analysis was more specific than ROI (ROI 42.1%,P=0.013).[Conclusion]:The curve-type distribution percentage with CAD tends to improve specificity significantly compared to the ROI method.Part IIValues of MRI-CAD in Differentiation between Benign and Malignant Breast Small Masses for Inexperienced and Experienced Readers[Purpose]:To investigate the additional value of the MRI computer-aided diagnostic (CAD) system in the evaluation of contrast-enhanced breast MRI for readers with different levels of experience.[Materials and methods]:A total of 235 small mass-like breast lesions (≤2.0 cm) seen on MRI were reviewed retrospectively. From each lesion detected either histological confirmation or follow up was obtained. All lesions were evaluated by four readers (2 inexperienced and 2 experienced) with and without use of the CAD system. All four readers diagnosed the lesions as either benign or malignant and scored a BI-RADS classification per lesion. Sensitivity, specificity and overall accuracy as measured by the area under the receiver operating characteristic curve (AUC) of both methods were compared for each reader. Interobserver variations were evaluated using kappa statistics.[Results]:There were 155 benign and 80 malignant small breast masses. All lesions were detected by all four readers. CAD improved specificity for both inexperienced and one experienced readers (55.5% vs.72.3% in reader 1; 58.1% vs.76.1% in reader 2; 72.3% vs.75.5% in reader 3), and statistical significance was found in both inexperienced readers (Preaderi,2<0.001,Preader3=0.404). Specificity in one experienced reader (reader 4) decreased with CAD used (74.8% vs.70.3%), but no statistical significance was found (P=0.265). Sensitivity in all four readers increased slightly by use of CAD, but there was no statistical significance (P reader 1-3= 1.000, P reader 4=0.480). The overall performance found in both inexperienced and one experienced readers (reader 3) improved significantly when using CAD (AUC,0.899 vs.0.947, P=0.002 in reader 1; AUC,0.839 vs.0.987, P<0.001 in reader 2; AUC,0.929 vs.0.971, P<0.001 in reader 3), but no significant difference was found for the most experienced reader with performance improved a!so(P reader 4=0.11). The level of agreement between the readers improved significantly when using the CAD method (P＝0.004).[Conclusion]:The CAD method tends to improve the diagnostic performance of reader with different levels of experience, especially the specificity of inexperienced readers. The use of CAD system is therefore recommended for inexperienced readers to avoid biopsy unnecessary.